Yeah, RAM and CPU are mostly meaningless for a switch, it's just for management purposes. The big question is if the switch chip is stable or if it has problems like the other CRS models.

Yeah, RAM and CPU are mostly meaningless for a switch, it's just for management purposes. The big question is if the switch chip is stable or if it has problems like the other CRS models.

I support a CRS312 that has 64 MB with L3 HW Offload. The switch reboots every 2 weeks when Memory gets maxed out.

So for me CRS312, CRS5xx series and this new one is Dead on Arrival (Aside of the 3X times price issue for additional port speeds (2.5 Gbps) and couple extra SFP+ and 2 x 40 Gbps that the Industry is abandoning.

CPU on a switch normally doesn't do a lot. The switch chip should do all the heavy lifting.

128M is certainly prohibitively low for running a sizable L3 network. For $1000 it should have an ARM with a 512MB at least.

I support a CRS312 that has 64 MB with L3 HW Offload. The switch reboots every 2 weeks when Memory gets maxed out.

So for me CRS312, CRS5xx series and this new one is Dead on Arrival (Aside of the 3X times price issue for additional port speeds (2.5 Gbps) and couple extra SFP+ and 2 x 40 Gbps that the Industry is abandoning.

So there's two things here.
On the CRS312, I can say that I've had one running for well over a year and the only reboots it's had are for ROS updates. It's only doing layer2 and "the snoopings"; routing is disabled and it's only "reachable" via a management VLAN.
I "feel like" there's a memory leak causing your reboots, but it's definitely not in any of the modules/features that I'm actually using. Throwing more RAM at a memory leak is an expensive bandaid... My recommendation would be to monitor RAM use and once you see it getting "rather high", generate a supout and put in a ticket with it so that their boffins can have a look.

The second thing, I definitely have mixed feelings on the switching/ports.
40Gbe ethernet is the same generation of tech that 10Gbe is... so for shops and use cases that are still "fat, dumb, and happy" with the 10Gbe, then 40Gbe is usually still ideal... because at the end of the day, those 40Gbe uplinks can be broken out to separate 10Gbe connections if more ports are needed, and on top of that (in this specific case) a single one of those 40Gbe uplinks is likely to be enough to keep the bulk of the 2.5Gbe connections happy.
Using a newer asic that has native 25/100Gbe switching, instead of 10/40Gbe switching, would have made this CRS326-4C+20G+2Q+RM even more expensive.

Precisely one of my points! and the price is 3X higher than the old CRS326-24G without any PoE and 1/5 of the memory!

Also, why is it using a mipsbe CPU when MikroTik is clearly moving everything they have to ARM? Even the $60 hAP ax lite has a dual core ARM CPU, but a $1000 switch does not?

In my opinion it's better to focus on real professional features (like advanced-switching, routing-protocols, encryption, API:s), unique hardware features (size, fanless, DC-fed) and affordability. People that want's the cheapest possible 2.5G/10G switch will head to Aliexpress.

It must be quite expensive to manufacture all that heatsink for nothing.

It's also very strange to have both fans and a big passive radiator. It's neither forced-air cooled nor passive-cooled. It must be quite expensive to manufacture all that heatsink for nothing.

Forced air devices have the advantage of be able to operate in warmer areas, passive ones can not.

If you are putting fans in the unit, why on earth why not design the whole switch then for a front-to-back or side-to-side airflow. If you are putting this device is a warm closet, the powersupplies and SFPs might be OK, but the chips will run very hot since the airflow doesn't pass the heatsink.